In case the rambling string of misspelled words and incoherent thoughts weren't dead giveaways, scientists have developed a method of machine learning to sniff out drunk tweets. Researchers from the University of Rochester collected 11,000 geotagged tweets over a year from two areas: New York and Monroe County, filtering the 140-character notes containing "drunk," "beer," "party" and other libation-related words. From there the school employed Amazon Mechanical Turks to decide whether the person sending the tweets was simply talking about booze or were actually drinking it while tweeting.
What's more, the team was able to develop a method for ascertaining if someone was drinking at home or from another location, with 80 percent accuracy. By matching alcohol-sales locations with the geotagged tweets the researchers discovered via algorithm that while more people in New York tweet about booze than in Monroe County, a higher number of residents of the former are drinking within 100 meters of their homes, if not in their homes outright. What's more, Monroe County residents drink more around a kilometer-plus distance from where they rest their heads.
MIT Technology Review notes that this methodology isn't a perfect predictor whatsoever, especially since Twitter skews younger and that "certain minorities" are overrepresented. But it is a whole lot less costly and faster than other research methods.The ultimate goal? Studying the ways that drinking changes based on age, sex and ethnicity in addition to hopefully heading off the tragic number of alcohol-related deaths.